Multimodal Attention Fusion for Target Speaker Extraction
This work provides an incremental improvement in robustness for audio-visual target speaker extraction systems, particularly for real-world applications where clue corruption is common.
This paper addresses the challenge of target speaker extraction in realistic scenarios where audio-visual clues may be corrupted. The authors propose a novel attention mechanism for multimodal fusion that weighs more reliable clues, resulting in a 1.0 dB improvement in signal-to-distortion ratio (SDR) over conventional methods on simulated data.
Target speaker extraction, which aims at extracting a target speaker's voice from a mixture of voices using audio, visual or locational clues, has received much interest. Recently an audio-visual target speaker extraction has been proposed that extracts target speech by using complementary audio and visual clues. Although audio-visual target speaker extraction offers a more stable performance than single modality methods for simulated data, its adaptation towards realistic situations has not been fully explored as well as evaluations on real recorded mixtures. One of the major issues to handle realistic situations is how to make the system robust to clue corruption because in real recordings both clues may not be equally reliable, e.g. visual clues may be affected by occlusions. In this work, we propose a novel attention mechanism for multi-modal fusion and its training methods that enable to effectively capture the reliability of the clues and weight the more reliable ones. Our proposals improve signal to distortion ratio (SDR) by 1.0 dB over conventional fusion mechanisms on simulated data. Moreover, we also record an audio-visual dataset of simultaneous speech with realistic visual clue corruption and show that audio-visual target speaker extraction with our proposals successfully work on real data.